Video surveillance systems have been widely used in various occasions, such as bank surveillance and public surveillance. The video surveillance system can momentarily track an object, such as a character or a vehicle, and thus play a very important role in the vision-based object tracking.
At present, the vision-based object tracking system is broadly divided into four categories. In the first category of vision-based object tracking system, the camera will continuously track the object in a field of view (FOV). However, when the object left the camera FOV and enters into the new FOV of another camera, the system will not be able to automatically track this object, or the system will be treated this object as a new object to continue tracking. The second category of vision-based object tracking system provides inter-camera FOV centralized tracking. However, as for the centralized vision-based object tracking system, it is not possible to reach a real-time tracking. In addition, when the number of tracked objects is increased, or the number of cameras in the system is increased, the system performance will be substantially reduced. The third category of vision-based object tracking system can across the camera FOV to track the object, but is seldom seeing on the market due to some factors. The fourth category of vision-based object tracking system requires additional sensors, such as radio frequency identification (RFID) sensors, to address the problem of re-identifying the object. This problem usually occurs in when the object across the camera FOV. However, such systems are not practical because of the need for additional sensors.
The current vision-based object tracking system does not satisfy the actual needs of the market, this invention provides a new intelligent distributed object tracking system to improve the problem of the current system, especially to improve the utility of the conventional vision-based object tracking system.
In addition, the conventional vision-based object tracking system also provides an event alert detection function. If a moving object triggers the alert condition, the system generates an alert event. However, the system cannot dynamically control whether or not only the objectives of specific to perform the alert detection, although it can obtain data of the specific object by querying the event database later.
Embodiment of the invention provides a distributed video surveillance system based on a handoff mechanism. When at least one object (human or vehicle) is under track, continuous tracking on multiple objects can be simultaneously performed. The video surveillance system according to embodiment of the invention performs multi-object tracking function even if the FOVs of multiple cameras are not overlapped. More particularly, the video surveillance system according to embodiment of the invention performs multi-object tracking function in the indoor/outdoor environment even if cameras are not calibrated in advance.